Renewable Energy Community Pairing Methodology Using Statistical Learning Applied to Georeferenced Energy Profiles

Abstract

Renewable energy communities (REC) are bound to play a crucial role in the energy transition, as their role, activities, and legal forms become clearer, and their dissemination becomes larger. Even though their mass grid integration, is regarded with high expectations, their diffusion, however, has not been an easy task. Its legal form and success, entail responsibilities, prospects, trust, and synergies to be explored between its members, whose collective dynamics should aim for optimal operation. In this regard, the pairing methodology of potential participants ahead of asset dimensioning seems to have been overlooked. This article presents a methodology for pairing consumers, based on their georeferenced load consumptions. A case study in an area of Porto (Asprela) was used to test the methodology. QGIS is used as a geo-representation tool and its PlanHeat plugin for district characterization support. A supervised statistical learning approach is used to identify the feature importance of an overall district energy consumption profile. With the main variables identified, the methodology applies standard K-means and Dynamic Time Warping clustering, from which, users from different clusters should be paired to explore PV as the main generation asset. To validate the assumption that this complementarity of load diagrams could decrease the total surplus of a typical PV generation, 18 pairings were tested. Results show that, even though it is not true that all pairings from different clusters lead to lower surplus, on average, this seems to be the trend. From the sample analyzed a maximum of 36% and an average of 12% less PV surplus generation is observed.